[Wien] MSR1a

[Laurence Marks]

As a followup to some recent questions about converging MSR1a, I wanted to
put a few comments out on the accuracy of a converged MSR1a calculation and
what it really is for everyone.

MSR1a is not something you will find (to my knowledge) in any other DFT
code, it is fundamentally different. Perhaps the closest is using Molecular
Dynamics in QE or similar to minimize the energy. While we now somewhat
understand it, there are still both theoretical and practical gaps in our
(my) understanding. Some of these gaps are quite large! Hence there are few
hard and fast rules.

It does not simply minimize the energy by moving the atoms, it also does
not simply converge the charge. Instead it finds a compromise. How close
the atom positions and density to the true minimum are depends upon the
system. We know empirically that it tends (at the moment) to work better
for insulators when it can converge very well. In some cases for metals the
convergence depends upon how the problem has been posed.

If you want energies etc at a 1 mRyd level and positional accuracy of 0.01
au it is probably fine in the default mode. These are normally larger than
the systematic errors due to inadequate functionals. If you have to go
beyond this you may want to check with PORT, which is often slower but well
understood and unconditionally convergent. This may not be needed, but
because of the gaps in understanding is a very safe approach. The default
mode is to switch to MSR1 at the end so you can look at the absolute forces
and estimate how well it is converged.

I am continuing to try to improve it, but it is not a simple linear
process. It involves intuitive guesses at what might work, then testing -
and most reasonable ideas fail. I am always interested in knowing cases
where it works well or badly, this helps me to understand it.

___________________________
Professor Laurence Marks
Department of Materials Science and Engineering
Northwestern University
www.numis.northwestern.edu
MURI4D.numis.northwestern.edu
Co-Editor, Acta Cryst A
"Research is to see what everybody else has seen, and to think what nobody
else has thought"
Albert Szent-Gyorgi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20141121/264ec784/attachment.html>